Experiments with Data-Intensive NLP on a Computational Grid

《Experiments with Data-Intensive NLP on a Computational Grid》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、ExperimentswithData-IntensiveNLPonaComputationalGridBadenHughes,StevenBirdHaejoongLeeEwanKleinDeptofComputerScienceLinguisticDataConsortiumSchoolofInformaticsUniversityofMelbourneUniversityofPennsylvaniaUniversityofEdinburghVictoria3010,AUSTRALIAPhiladelp

2、hiaPA19104,USAEdinburghEH89LW,UKfbadenh,sbg@cs.mu.oz.auhaejoong@ldc.upenn.eduewan@inf.ed.ac.ukAbstractthecommunityneedstoexplorenovelwaystoLargedatabasesofannotatedtextandspeecharewidelycollaborateontheprocessingoflargedatasets.usedfordevelopingandtesting

3、languagetechnologies.How-ever,thesizeofthesecorporaandassociatedlanguagemod-Thispaperreportsaseriesofexperimentselsareoutpacingthegrowthofprocessingpowerandnet-withdata-intensivenaturallanguageprocessingworkbandwidthavailabletomostresearchers.Thesolu-onac

4、omputationalgrid.Acomputationalgridtion,webelieve,istoexploitfourcharacteristicsoflanguagetechnologyresearch:manylargecorporaarealreadyheldfacilitateslarge-scaleanalysisusingdistributedatmostsiteswheretheresearchisconducted;mostdata-resources.Giventheprev

5、alenceoflargedataintensiveprocessingtakesplaceinthedevelopmentphasesourcesinnaturallanguageengineeringandandnotatrun-time;mostprocessingtaskscanbeconstruedasaddinglayersofannotationtoimmutablecorpora;andtheneedforrawcomputationalpowerinthemanyclassesoflan

6、guagemodelscanbeapproximatedastheanalysisandmodellingofsuchdata,thegridsumofsmallermodels.Wereportonaseriesofexperimentscomputingparadigmprovidesefficienciesandwithdata-intensivelanguageprocessingonacomputationalgrid.Keyfeaturesoftheapproachareitsuseofascr

7、iptingscalabilityotherwiseunavailabletonaturallanguageforeasydisseminationofcontrolcodetoprocess-languageengineering.Agridbrokeractsasingnodes,theuseofagridbrokertomanagetheexecutionoftasksonremotenodesandcollatetheiroutput,theuseaatask-farmingengine,empl

8、oyingthesimplestdata-decompositionapproachbywhichparametricandparal-formofdistributioninwhichparalleltasksarelelprocessingofindividuallanguageprocessingcomponentsfullyindependent.Thuswedonotenvisageoccursonsegmented